Ensemble Deep Learning Framework Combining CNN, Transformer, and XGBoost for EEG-Based ADHD Detection

Attention Deficit Hyperactivity Disorder (ADHD) is one of the most common neurodevelopmental diseases affecting children and adolescents, frequently causing difficulties with attention, impulse control, and cognitive performance. Early and correct ADHD diagnosis is critical for successful intervention and treatment planning. EEG recordings are valuable biosignals in determining the presence of this disorder. However, previous studies using conventional machine learning algorithms have failed to capture the complexity and nuances of these bio-signals, leading to less than optimal results. The current study describes an ensemble-based deep learning system for identifying ADHD using EEG signals. The proposed approach uses convolutional neural networks, transformer networks, and extreme gradient boosting methods to automate ADHD diagnosis based on EEG data. The system makes use of EEG signals collected from 19 channels, which were pre-processed and normalized before being used in the model. The convolutional neural network was used to extract local features from EEG data, while the transformer network captured global information from encoded input signals. Finally, the extreme gradient boosting classifier was utilized to divide the collected features into two groups: ADHD and normal. The model's performance was assessed on a benchmark EEG dataset for ADHD classification. The model had an accuracy of 98.94%, 99% precision, recall, and F1-score, suggesting good performance in identifying the two types of EEG signals. The confusion matrix showed few classification mistakes, and a comparison of findings demonstrated that the suggested CNN-Transformer-XGBoost framework outperformed existing machine learning and deep learning models in distinguishing ADHD from normal EEG signals. Thus, the suggested framework can be used as a computer-aided diagnostic system to help doctors identify ADHD patients by analyzing EEG records.